Neil Batlivala

Solve for information asymmetries! Architecture & Civil Engineering - Layout interpretation ~other industries: Healthcare - <Prescription> interpretation Financial Products - <Subject to market risks documentation> interpretation Loans - <Fine print> interpretation There’s going to be a slew of startups commercialising this year with a value prop of simplification using AI

5

🎬 Matt Damon and 🏈 Colin Kaepernick have invested in it, and so has 💰Andreessen Horowitz... Function Health's executive team has "engineered a first-of-its-kind approach that focuses on prevention and empowerment"—very much in line with what we aimed to achieve with Carebook when we started it over six years ago. Could they be the ones to fulfill their lead investors' prophecy (link at the bottom) and Dr. Eric Topol, MD's premonitions in his books "The Patient Will See You Now" and "Deep Medicine"? They have just secured a $50 million grant. Let's hope this is not another Theranos and that we are truly witnessing the rise of the next generation of direct-to-consumer health companies. These companies will be key players in the "transition from reactive to proactive" health care that the system—and we—desperately need. What do you think? Links: 📝 Investment round press release: https://lnkd.in/eUG87qkf 🔗 Andreessen Horowitz's prophecy: https://lnkd.in/eZATCf4f

4

Where are healthcare providers? Early this week, I had the random idea of " geocoding" the entire NPPES data, i.e., retrieving latitude and longitude coordinates from the address fields and providing the data to the minilabs users. Why am I doing this? Well, I can think of a few reasons. First, the address fields are surprisingly unstructured; the same address can have many representations. If we transform these free texts into physical coordinates, we can easily calculate the distance of two addresses. Second, business names may change over time, but not their physical coordinates. Identifying the same practices with this additional data point would be more manageable. Finally, we humans are visual animals; provider clusters are easier to spot on a map rather than on an Excel file. Yes, a picture is worth a thousand words. Although I am still far from geocoding the full NPPES dataset, I wanted to share my interim progress. The map below shows the geocoded addresses near Atlanta, GA. I used Hex to visualize the heatmap. What do you see on the map? I see two clusters near Decatur and Sandy Springs, i.e., two red blobs. They are Emory University Hospital and Northside Hospital. I didn't realize so many providers are clustered around these two hospitals, and yes, seeing this on the map changed my view. I can't wait until the first pass of the NPPES geocoding is done. I plan to improve the accuracy of the geocoding results as well. Stay tuned! #censusgeocoding #opendata #providerdata #nppes #healthcareanalytics

79

31 Comments

😅 How did Dan end up manufacturing large quantities of pharmaceutically optimized LSD? 1. Founder of HealthMode 2. Acquired by MindMed to become CMO 3. Run an FDA breakthrough program for LSD in anxiety Full episode in the comments 👇

13

1 Comment

Building on legacy systems is a huge problem for health startups. Historically, health has been much harder for software to disrupt than other industries. Healthcare is 20% of the US economy, but only one of the largest 100 public software companies is a healthcare company. Now, AI is leapfrogging the outdated tech in health. Providers don’t want new software, but they definitely do want more people. The promise of AI is that it can automate a whole service, rather than just relocating the same work to a shinier interface. The only problem is that these companies have to provide automation on top of layers of legacy tech. How do you automate a process when there aren’t good APIs to connect to? Thats why many AI Health companies use CloudCruise. We provide the AI hands which can operate software (writing notes or extracting data) on behalf of AI brains.

44

3 Comments

I’ve been playing around with the new version of ChatGPT (o1 preview). It’s pretty impressive. For example, it can correctly count how many Rs there are in “strawberry”. Doesn’t seem like a big deal to you and me, but that kind of task was a big limitation in LLMs reasoning capabilities until now. Something else it can now do is below. Previously, no LLM I tried (and I tested most) could perform this simple task. There would never be simply six circles in a square. In the image below, the output from version 4o and the new o1 are presented. While the image in the older version is more refined, the output from the new model is accurate, showing a significant leap in its abilities.

15

2 Comments

I am torn by this as well. When I see a new platform shift, I see the opportunity for a shift that benefits the end user. The ultimate end user in healthcare is the patient. In the current model, the structure is to chase revenue by driving it up or maximizing profit. The AI platform shift, targeting low-hanging fruit like administrative efficiencies, follows this path. It's reminiscent of the famous quote: "We wanted flying cars, but we got 140 characters." The same thing is happening all over again, without truly challenging the status quo. Maybe AI can help people get less sick, prevent disease, and navigate spending less money in the first place. But in the current model, it's difficult to get penetration. Ultimately, everyone needs short-term metrics and traction to raise another round of funds. I am not criticizing those who play this game. I respect them for playing it well. However, they shouldn't position themselves as lowering costs for patients by decreasing administrative costs. The trickle-down effect simply doesn’t work.

4